zhengjun-ai/moirephotorestoration-mcnn

This is an unofficial reproduction of paper Moir´e Photo Restoration Using Multiresolution Convolutional Neural Networks.(PyTorch)

moirephotorestoration-mcnn's Issues

您好请教一下数据预处理部分

您好，我下载了testdata，发现每个图片都有一个黑色“相框”用作对齐，但是在您的代码中没有“对齐”的代码，请问您是怎么理解和处理的呢？多谢指教

No access to Google Drive download link

Hello!

I started downloading the dataset, but access to the dataset has been closed since last week when following the link to google drive. I also requested access via this link, but a week later I still have not received a response.

The data download link

The baidu network disk link does not work, Can you provide a new baidu nework dist link?

TIP2018链接失效

如题，链接中的TIP2018的数据集链接失效了

摩尔纹生成

您好，请问一下您有摩尔纹模拟的python代码？我最近也在做相关研究，想尝试模拟一下摩尔纹，但似乎没找到开源的代码。

一个比较严重的模型代码错误

你好，我注意到在def conv_block这个函数里，不能用初始化一个卷积核到列表里再乘5，因为这样你5个卷积核是共享参数的，事实上是只有一个卷积但网络循环经过了五次。
这也是为什么我测参数量时发现不对，多次debug发现了这个漏洞。
或许更改一下这里就可以复现出report的结果。

祝好。

dataset

兄弟，数据集有新的链接吗

请问这个项目可以用来检测一张图片是否包含摩尔纹吗？

看了您的博客过来的

Questions about the dataset

Thanks a lot for sharing the dataset and code.
I've noticed that in most paper[1,2,3] introductions, it is mentioned that the TIP2018 dataset contains 135,000 images with a resolution of approximately 400x400. However, the dataset downloaded from the Baidu Cloud link has a smaller number of images than 135,000 and a resolution of 800x800. Is this normal?

1. Real-Time Image Demoireing on Mobile Devices. ICLR2023.
2. Towards Efficient and Scale-Robust Ultra-High-Definition Image Demoireing. ECCV2022.
3. FHDe2Net: Full High Definition Demoireing Network. ECCV2020.

Pretrained model weights missing

Hi @ZhengJun-AI,

Can you provide pretrained model weights?